Neural CAD

It is thought that the term neural CAD was first used (in relation to a specific commercial application) by Autodesk during their Autodesk university event in September 2025, in particular Mike Haley, leader of the machine intelligence group at Autodesk Research.

Neural CAD refers to a new category of generative AI foundation models which are designed to directly reason with and create CAD geometry through prompts. Autodesk defines this as where AI models are trained to understand and generate editable Boundary Representations (BREP) geometry, which are ready to edit parametric models rather than static forms, from text prompts.

The work moves beyond simple Large Language Model (LLM) integration into completely new CAD engines, where by initial base models can be created from prompts such as 'design a 5 storey building' and the creation is then an editable 3D CAD model as opposed to a 3D form or image. see AU 2025 The Design & Make Conference

In his blogpost 'upcoming 3D generative AI foundation models for Autodesk Fusion and Forma' (3 September 2025) Mike Haley says

"In the early stages of a design project during ideation and conceptualisation, tools should accommodate the broadest methods of expression. Early specifications, outlines, rough-sketches, and photos, among others, are all essential in expressing ideas. Creators need technology that does not obstruct their creative flow. Instead, tools should enable rapid and iterative exploration of ideas, regardless of how detailed or approximate those ideas might be. Over the past decade, research has underscored that such early-stage explorations significantly enhance project metrics like cost, sustainability, and suitability. Not only have these flexible tools been lacking but attempts to develop them have resulted in solutions that cannot be transferred easily to detailed design development..."

"Fast-forward to today, and we are in a transformative era in design and make technology. Picture a scenario where your computer comprehends spoken language, sketches, three-dimensional design data, and industry-specific workflows. Now, enrich this scenario with decades of your team’s project knowledge."

"Autodesk is pioneering the development of a range of neural AI foundation models focused on design to make problems. The first of these is neural CAD, a category of generative AI models trained to directly reason both about CAD objects and industrial and architectural systems. Moving far beyond the practice of pairing large language models with existing solutions, Autodesk’s approach completely reimagines the traditional software engines that create CAD geometry or translate between industrial and architectural systems. Neural CAD foundation models offer a path to enabling the kinds of explorative interactions mentioned above but also promising a future of more connected design and make processes."